TSMC’s $100 Billion Arizona Bet Still Lacks New Money

TSMC’s first Arizona fab has been in high-volume production since the fourth quarter of 2024; its second is complete, with output expected in the second half of 2027, and construction of a third has begun.^{[1, 2]} That operating record gives substance to the company’s American expansion. It does not, however, establish that the newly advertised $100 billion will alter TSMC’s earnings power: the latest announcement calls the amount incremental, while earlier reporting says the wider package already counted $100 billion pledged in 2025.^{[3, 4]}

That distinction matters more than the $265 billion total. A genuinely new capital allocation would enlarge TSMC’s planned US manufacturing and packaging base beyond the path already embedded in prior commitments; a repackaged figure would mainly strengthen the political presentation of spending that was already expected. The announced plan adds four advanced facilities and takes the intended US footprint to 12 leading-edge manufacturing and packaging sites, but TSMC has not published a timetable that ties those additional plants to commercial production.^{[2, 3]}

The physical build-out is nevertheless moving. Tool installation at the second fab was scheduled for 2026, the third fab is under construction, and TSMC is seeking permits for a fourth fab and its first advanced-packaging facility in Arizona.^[1] The first fab’s production proves that TSMC can move an Arizona project from commitment to output.^[2] It does not prove that the next wave will arrive on the announced scale, at the required speed, or with acceptable returns.

Execution in Arizona has been proved, but the timetable and returns remain uncertain

Time is the first pressure point. Earlier analysis placed the newly announced fabs beyond 2030, with a possible rollout extending to 2035, because the third confirmed fab itself was not expected to reach production before the end of the decade.^[5] Capital spending of this kind unfolds over years rather than in one budget cycle, and TSMC’s record $45 billion package for 2026 covers global demands as well as the Arizona build-out.^[6] The earnings effect therefore depends less on the headline commitment than on when each facility receives tools, qualifies processes, secures customers, and reaches usable output.

Demand is not the obvious weakness. TSMC says global AI-related demand remains “extremely robust”, and high-performance computing generated 55% of fourth-quarter revenue, up from 40% in the same quarter of 2022.^{[7, 8]} HBM shipments are projected to keep rising through 2027, while CoWoS capacity, advanced-node pricing, and hyperscaler capital expenditure continue to expand.^[9] Those indicators support the case for more advanced-node and packaging capacity. They do not determine whether Arizona is the most profitable place to provide it.

That question turns on an economic structure TSMC has not disclosed at facility level. Estimates of the US cost penalty vary sharply: one account cited a 150% premium to build an initial fab at a new site, while another estimated that processing a 300mm wafer in Arizona costs less than 10% more than in Taiwan because equipment dominates wafer economics and labor represents less than 2% of total cost.^[10] Subsidies can narrow the gap, but large recipients may face conditions including profit-sharing when returns exceed projections.^[11] Without Arizona revenue, margin, and return data, production validates execution, not financial success.

Strong demand cannot resolve the undisclosed economics of US production

Infrastructure can widen that gap before a wafer reaches a customer. Arizona’s water supply remains a major concern, and TSMC is also seeking reliable power while navigating environmental and electricity-consumption rules.^[2] The company has described water-treatment and recycling plans, but the expansion also depends on public infrastructure and regulatory accommodation.^[2] Those constraints can delay tool installation, raise operating costs, and push revenue further beyond the period in which capital is committed.

The regional ecosystem is further advanced than it was when TSMC selected Arizona. Since 2020, the state has attracted more than $210 billion in semiconductor-related investment and more than 25,000 projected jobs.^[12] TSMC also wants Taiwanese chemical and equipment suppliers to follow it, although those moves may require regulatory changes.^[2] Suppliers and workers could benefit before TSMC does because construction, equipment orders, and hiring can occur well before a new fab reaches efficient utilization. Yet projected employment and ecosystem spending remain weaker proof than signed supplier awards, completed hiring, and sustained production.

Geographic diversification has strategic value even if it dilutes near-term margins. More US capacity could give customers a qualified second manufacturing site and reduce dependence on a single location, a recognized mechanism for improving supply-chain resilience.^[13] But TSMC still concentrates more than 90% of manufacturing capacity in Taiwan; 87% of its employees work there, and close to 90% of its global workforce is Taiwanese, including 88% of managers.^[14] Arizona can add redundancy without quickly reproducing the engineering density, supplier network, and managerial depth of TSMC’s home base.

Strategic diversification adds resilience without replicating Taiwan’s manufacturing depth

Nor does the expansion yet establish a competitive loss for Samsung or Intel. TSMC holds about 72% of the foundry market and an estimated 95% of leading-edge production, while Samsung has struggled to match its 3nm performance and Intel’s 18A ramp remains largely directed at internal products.^[15] Additional Arizona capacity may reinforce that position, particularly if it reaches 2nm-class production and advanced packaging at scale.^{[4, 15]} But no disclosed customer shift, supplier reallocation, or canceled rival project ties the US announcement to an observed change in market share.

The capital-allocation question can be tested before the new plants produce a single wafer. A TSMC filing that reconciles the 2025 pledge with the latest $100 billion announcement would establish how much spending is truly new; permits and construction starts for the fourth fab and first advanced-packaging plant would show whether the four-facility promise is becoming executable.^{[1, 2, 4]} A facility-by-facility schedule, evidence of 2nm-class production in Arizona, named customer capacity commitments, and annual US capital expenditure above the prior path would move the case from intention to operating visibility.^{[4, 5]}

Until then, the proven change is political and industrial: TSMC has enlarged its stated American commitment while continuing to build in Arizona under strong AI demand.^{[1, 2, 3]} The unproven change sits on TSMC’s balance sheet. It will appear only when the company shows that the new money is genuinely incremental, the facilities are commercially loaded, and Arizona can earn enough to justify capital that might otherwise have remained inside Taiwan’s more complete manufacturing system.^{[2, 4, 14]}